The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of voice agents that process untrusted data from multiple sources, creating a surface for indirect prompt injection attacks.
Ingestion points: The agent ingests data through user audio transcripts (via onUserTranscript in the JavaScript SDK) and external tool responses (the result field returned by webhooks).
Boundary markers: The provided system prompt examples do not include delimiters or instructions to ignore embedded commands within user speech or tool outputs.
Capability inventory: The agents possess significant capabilities including making outbound phone calls (outbound_call), transferring calls to external numbers or other agents, and executing server-side webhooks.
Sanitization: There is no evidence of sanitization or validation of external tool results or transcripts before they are incorporated into the LLM context.
[EXTERNAL_DOWNLOADS]: The skill references and installs official developer tools and libraries from the ElevenLabs platform.
Downloads the command-line interface using npm install -g @elevenlabs/cli.
References the official conversation widget script from https://unpkg.com/@elevenlabs/convai-widget-embed.
Downloads SDKs for Python (elevenlabs) and JavaScript (@elevenlabs/elevenlabs-js).
[COMMAND_EXECUTION]: The instructions involve executing shell commands for project initialization and authentication.
Uses elevenlabs auth login to handle API credentials.
Uses elevenlabs agents push and pull to synchronize local configurations with the remote platform.

agents